Illustrative example. This memo was written in October 2026, after the outcome was known, to show how a desk memo works, including a miss. Its pre-event sections use only information public on 13 March 2023. It is not part of the desk’s track record and is excluded from the hit rate.
Hypothesis
Silicon Valley Bank was closed on 10 March 2023 and Signature Bank on 12 March. On 12 March, US authorities said all depositors of both banks would be protected. The Federal Reserve also created the Bank Term Funding Program, which lends against eligible securities valued at par.
The memo argues that this is not enough. Depositors at other banks with large uninsured balances now have a reason to run first. Under the BANKING_PANIC archetype, runs spread until authorities guarantee all uninsured deposits.
Contrary reading
Under INFLATION_RATE_SHOCK, the losses come from a rate rise that hit some balance sheets far harder than others. If so, stress stays concentrated in banks that share SVB’s profile and works out over months.
Base rate
Greenwood, Hanson, Shleifer and Sørensen (2022) find a 3-year crisis probability of about 7% in normal times. The 30% conditional figure is this desk’s own judgment for a system-wide panic in this setup. It is not a result from the study.
Falsification criteria
- No explicit guarantee of all uninsured US deposits is announced by 30 June 2023.
- Two or fewer further banks appear on the FDIC failed-bank list with closing dates from 13 March to 30 June 2023 inclusive, not counting SVB and Signature.
The threshold in criterion 2 is the desk’s judgment, set before the event. A system-wide panic would close several more banks, not one or two.
Post-mortem (30 June 2023)
- Outcome: invalidated. Both criteria were met.
- What happened: no blanket guarantee was announced. Only one further US bank failed in the window: First Republic Bank, closed and sold to JPMorgan Chase on 1 May 2023.
- Likely cause of the miss: the memo treated the new facility as a stopgap. Lending at par meant banks did not have to sell securities at a loss, which plausibly blunted the fire-sale channel a broad panic needs. The outcome is consistent with this explanation but does not prove it.